| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1039.1 | | NABETH::alan | Dr. File System's Home for Wayward Inodes. | Thu Apr 10 1997 11:34 | 9 |
| People have managed to use News on AdvFS by arranging to have
enormous amounts of metadata preallocated for the purpose of
supporting the necessary number of files. Others, or in addition
to this, regularly defragment their file systems. The later
takes considerable maintenance. The former takes upfront
educated guess work.
The fundemental problem seems to be one of design limitations
and not bugs, which makes it hard to "fix".
|
| 1039.2 | Large extent count with 90% performance indext? | NNTPD::"[email protected]" | Feynman Lo | Mon Apr 14 1997 20:09 | 52 |
| I wonder why is it possible to have a performance index of 91% whereas
the extentCnt of the file is 1175!
The file domain contains only one single LSM volume on RAID disks.
In this case, is it necessary to defragment the file?
This occurs in the news history file of an ISP.
#showfile -x /newslib/history
Id Vol PgSz Pages XtntType Segs SegSz Log Perf File
23e.8005 1 16 18498 simple ** ** off 91% history
extentMap: 1
pageOff pageCnt vol volBlock blockCnt
0 2 1 38946592 32
2 1 1 38950496 16
3 1 1 38950528 16
4 1 1 38950576 16
5 1 1 38950624 16
6 1 1 38950656 16
7 1 1 38950688 16
8 2 1 38950736 32
10 2 1 38950816 32
12 3 1 38950880 48
15 3 1 38950976 48
18 4 1 38951072 64
:
:
:
18396 6 1 7195152 96
18402 1 1 7370336 16
18403 15 1 7370384 240
18418 14 1 7503856 224
18432 2 1 7504720 32
18434 16 1 7570352 256
18450 3 1 7612304 48
18453 2 1 7613824 32
18455 9 1 7613968 144
18464 2 1 7614208 32
18466 16 1 7728432 256
18482 7 1 7790464 112
18489 9 1 7790800 144
extentCnt: 1175
- feynman
[Posted by WWW Notes gateway]
|
| 1039.3 | so far so good | NNTPD::"[email protected]" | Jeff Dickens | Fri Apr 18 1997 11:07 | 14 |
| I'm running the latest advfs patches for V4.0b, and I removed each volume
in my spool area domain one at a time and re-added them with -x 2048 -p 57600.
It's been 5 days and it's holding up fine.
I wrote a script to do a "ctlinnd pause", then a defrag and balance and
then a "ctlinnd go", and it's working ! It only took about 3 hours last night
for a 60% full 12GB domain. And I have only 2 BMT extents per volume.
I highly recommend the sys_check script to analyse advfs performance,
among other things.
[Posted by WWW Notes gateway]
|